Spatial consistency and bias in avalanche forecasts – a case
study in the European Alps
Abstract. In the European Alps, the public is provided with regional avalanche forecasts, issued by about 30 forecast centers throughout the winter, covering a spatially contiguous area. A key element in these forecasts is the communication of avalanche danger according to the five-level, ordinal European avalanche danger scale (EADS). Consistency in the application of the avalanche danger levels by the individual forecast centers is essential to ensure the greatest value for users, particularly those utilizing bulletins issued by different forecast centers. As the quality of avalanche forecasts is difficult to verify, due to the categorical nature of the EADS, we investigated forecast goodness by focusing on consistency and bias exploring real forecast danger levels from four winter seasons (477 forecast days). We qualitatively describe the operational constraints associated with the production and communication of the avalanche bulletins, and we propose a methodology to quantitatively explore spatial consistency and bias. We note that the forecast danger level agreed significantly less often when compared across national and forecast center boundaries (about 60 %), as compared to within forecast center boundaries (about 90 %). Furthermore, several forecast centers showed significant systematic differences towards using more frequently lower (or higher) danger levels than their neighbours. Discrepancies seemed to be greatest when analyzing the proportion of forecasts with danger level 4-High and 5-Very High. Operational constraints in the production and communication of avalanche forecasts, such as the size of warning regions, as well as differences in avalanche winter regimes, and variation in the ways the EADS is interpreted locally may contribute to inconsistencies. All these issues highlight the need to further harmonize the forecast production process and the way avalanche hazard is communicated to increase consistency, and hence value for the user.